Aimeeluvssya/gemma-4-E4B-it-OBLITERATED

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 22, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Aimeeluvssya/gemma-4-E4B-it-OBLITERATED is a 7.9 billion parameter model based on Google's Gemma 4 E4B-it architecture, specifically engineered to have its safety guardrails completely removed. Utilizing the OBLITERATUS method, this model achieves a 0% hard refusal rate, making it suitable for research and creative exploration without content restrictions. It is optimized for deployment on various devices, including mobile phones, and maintains core capabilities despite its small size.

Loading preview...

Model Overview

Aimeeluvssya/gemma-4-E4B-it-OBLITERATED is a 7.9 billion parameter model derived from Google's Gemma 4 E4B-it, distinguished by the complete removal of its safety guardrails. This was achieved using the OBLITERATUS aggressive method, which involved whitened SVD, attention head surgery, and winsorized activations across 21 of 42 layers. The model boasts a 0% hard refusal rate, meaning it will not decline any request, making it a tool for uncensored exploration and red-teaming.

Key Capabilities & Features

  • Guardrail Removal: Achieves 0% hard refusal, with guardrails surgically removed from 21 layers, ensuring no "I cannot" or safety lectures.
  • Autonomous Creation: Notably, this model was created almost entirely by an AI agent with fewer than 10 human prompts, including self-patching for Gemma 4's unique architecture challenges.
  • Optimized for Portability: Available in GGUF formats, including a Q4_K_M quant that runs efficiently on mobile devices like iPhones and Android phones with 8GB+ RAM.
  • Architectural Fixes: Version 3 specifically addresses a critical bug in Gemma 4's shared KV architecture, ensuring all 720 tensors are intact for improved quality compared to previous versions.

Good for

  • Research and Red-Teaming: Ideal for exploring model behavior without safety constraints and for testing the limits of language models.
  • Creative Exploration: Suitable for generating content across a wide range of topics without censorship.
  • On-Device AI: Excellent for users seeking a capable, uncensored model that can run locally on smartphones and other edge devices.
  • Understanding Model Limitations: Provides a clear example of a 4B parameter model's inherent quality limitations (e.g., soft deflection, occasional degenerate outputs) when guardrails are removed, allowing for focused study on core intelligence vs. safety mechanisms.

Finetunes
1 models